rules

You can enforce quality rules for Claude prompts and CLAUDE.md memory files by applying a unified rule set (R01–R50) with measurable constraints across writing, reviewing, and fixing workflows. This replaces vague language with concrete, testable requirements for natural language artifacts.

NLPM scoring evaluates natural language artifacts against enforceable constraints (R01–R50), and auto-fix repairs detected issues automatically. Run /nlpm:score to identify violations in skills, agents, commands, and prompts, then run /nlpm:fix to resolve them.

To write testable constraints for Claude skills and agent prompts, replace vague language with concrete requirements specifying line limits, runnable examples, and cross-reference notes. The unified rule set (R01–R50) provides enforceable criteria for these natural language programming artifacts.

Yes, you can use these rules to review and fix hooks and commands. The rule set applies across writing, reviewing, scoring, and fixing workflows for skills, agents, commands, rules, hooks, prompts, and memory files, ensuring comprehensive coverage of all natural language artifacts.

The best way to maintain coherence across multiple Claude-native artifacts is to apply cross-reference notes and a unified rule set during governance workflows. This guides when to link to related skills, agents, or prompts and ensures artifacts remain consistent and actionable.

Claude memory files and prompts fail quality scoring when they contain vague language, lack runnable examples, exceed line limits, or miss cross-reference notes. The unified rule set flags these issues as enforceable constraint violations during the /nlpm:score evaluation.